IPB201: The Never-Ending Prefix Debate: Revisiting Best Current Practices
Hosts of the IPv6 Buzz podcast discuss a draft RFC by Jordy Martinez proposing updated best current practices (BCP) for IPv6 prefix usage. They cover point-to-point link addressing, the pros and cons of Global Unicast vs. ULA vs. Link Local addressing, and service provider prefix allocation conventions. The episode argues that many old IPv6 rules of thumb are outdated and need revision.
Summary
The episode centers on a draft RFC authored by Jordy Martinez that aims to establish updated Best Current Practices (BCP) for IPv6 prefix usage. The hosts — Ed Horley, Nick Boraglio, and Tom Coffin — frame the discussion around the need to modernize outdated IPv6 conventions that have persisted in the industry for over a decade.
The first major topic is point-to-point link addressing. Historically, /64 prefixes were mandated by SLAAC requirements, and later /127s were recommended via RFC 6164 to address the 'ping pong attack' and neighbor cache exhaustion issues. The hosts argue that these hardware-based vulnerabilities are no longer relevant, as modern equipment has addressed them in code. They suggest that allocating a /64 and optionally assigning a /127 from within it offers the best flexibility, allowing future changes without renumbering. The use of /126s, /112s, or /120s is dismissed as unnecessary complexity driven by legacy habits.
The second topic covers the choice between Global Unicast Address (GUA), Unique Local Address (ULA), and Link Local addressing. The hosts express strong preference for GUA in almost all cases. ULA is described as frequently misused — particularly the habit of sequentially numbering prefixes (FD00, FD01, etc.) rather than using random global IDs as the RFC intended — which defeats its aggregation and uniqueness benefits. The hosts also note that ULA causes problems in dual-stack environments due to RFC 6724 address selection behavior. Link Local-only routing (per RFC 7404) is acknowledged as technically functional but operationally problematic due to reduced visibility and path MTU discovery complications.
The third topic addresses service provider prefix allocation practices. The hosts discuss two design philosophies: using a dedicated pool of /64s for PE-to-CE point-to-point links separate from customer prefixes, versus aggregating the point-to-point link within the customer's allocated prefix block (e.g., a /48). Both approaches are presented as valid. The hosts also strongly advocate for stable, non-churning prefix delegations to customers, regardless of size, noting that prefix instability causes real operational problems. They criticize large U.S. carriers for assigning only /64s to customers rather than more generous allocations like /48s.
Finally, the hosts touch on interface numbering conventions, with a preference for using hexadecimal values like 'A' and 'B' rather than '1' and '2' for point-to-point links, as A/B works consistently whether using /127 or /64 prefixes. They also note that embedding operational metadata (like VLAN IDs or router IDs) into address space is an operational convention that tends to break down quickly in practice and should not be treated as a hard rule. The episode closes with an encouragement for community members to engage with the draft RFC while it is still open for comment.
Key Insights
- Nick Boraglio argues that the neighbor cache exhaustion and ping pong attack issues that motivated /127 recommendations are no longer relevant because modern equipment has solved them in software, making the strict use of /127s an outdated convention rather than a necessary security measure.
- The hosts contend that ULA is widely misused in practice because operators sequentially number prefixes (FD00, FD01, etc.) rather than using random global IDs as specified, which eliminates the aggregation and uniqueness benefits ULA was designed to provide.
- Nick Boraglio claims that in dual-stack environments, ULA is more problematic than GUA due to RFC 6724 address selection behavior, which can cause traffic to fall back to IPv4 unexpectedly rather than using the ULA IPv6 address.
- The hosts argue that Link Local-only routing (RFC 7404) is technically functional but operationally problematic because it creates a 'black box' with no end-to-end visibility and complicates path MTU discovery, making troubleshooting significantly harder for network operators.
- Nick Boraglio asserts that embedding operational metadata such as VLAN IDs or router IDs into IPv6 address space as a naming convention breaks down quickly in practice because a single exception invalidates the entire convention, making IPAM automation a more reliable alternative.
Topics
Full transcript available for MurmurCast members
Sign Up to Access